Idea 08/Neuro & BCI/Toxicology platform/3 min

The Connectome Digital Twin

A complete fly nervous system is now a wiring diagram. Wiring diagrams are screening instruments if you simulate them.

With all 166,700 neurons and 124 million synapses of the male fly mapped, sublethal neurotoxicity can be predicted from circuit dynamics instead of from lethality, turning the connectome into a regulatory assay.

01What changed

In September 2026 a consortium mapped every connection in the brain and ventral nerve cord of an adult male fly: 166,700 neurons, 124 million synapses, from 66 tissue pieces imaged by seven electron microscopes for a year.1 UKRI, Cell, September 2026
The same dataset localised sex-specific wiring: about 95 percent of cells are shared between sexes, and the remaining 5 percent differ enough to plausibly drive distinct behaviours.2 Google Research, 2026

02The gap

Connectomes are published as anatomy and used for tracing. They are almost never used as the substrate of a predictive model for chemical perturbation, even though the whole point of a wiring diagram is that it constrains dynamics.

03The idea

Instantiate the olfactory and antennal-lobe circuit as a spiking network whose connectivity is taken from the connectome rather than tuned. Keep the model deliberately small: a few thousand neurons around the projection-neuron to Kenyon-cell pathway, with the rest held as fixed input statistics.

Then dose it. For each compound, encode a plausible mechanism, for example inhibition of a specific ion channel or disruption of a neurotransmitter reuptake transporter, at concentrations that map to measured exposure. Simulate the circuit response to odour plumes and extract behavioural proxies: discrimination index, response latency, and the slope of the dose-response.

The prediction that matters is a threshold, not a direction: the concentration at which odour discrimination degrades by a defined amount. That is a number a regulator can compare with an environmental exposure, and it is exactly what current assays do not produce.

Calibrate against real Drosophila behaviour at three concentrations for ten compounds, then predict five held-out compounds blind. A model that gets the ordering right is useful even if absolute thresholds are off by a factor of two.

04Why now

The complete CNS connectome is three months old, simulation tooling for spiking networks is mature, and pesticide neurotoxicity screening is under pressure to move away from vertebrate tests. The three trends meet at this dataset.

05The decisive test

Do this

Build a connectome-constrained olfactory model, calibrate on ten compounds, and blind-predict behavioural thresholds for five more.

Success looks like

Rank correlation above 0.7 between predicted and measured thresholds across the blind set, with the model beating a generic randomly wired network of the same size.

Kills it

If a randomly wired network matches the connectome-constrained one, wiring specificity is not carrying the information and the twin adds nothing.

06What would break it

Connectome data are from one individual and omit synaptic weights and neuromodulation.
Mapping molecular mechanism to model parameters may be underdetermined.
Fly neurotoxicity may not extrapolate to pollinators, which is the actual policy question.

07If it works

Pesticide screening gains a mechanistic, non-animal tier that predicts sublethal harm, and connectomes become instruments rather than atlases.